[BUG] voice package is not being used - `voice.GroqTranscriber` is not part of the process any more

[dim]

Quick Summary

Voice messages are broken since v0.2.0. I don't see any more references to voice.NewGroqTranscriber or to the voice package in general anywhere in the code.

Environment & Tools

• PicoClaw Version: v0.2.0
• Go Version: n/a
• AI Model & Provider: n/a
• Operating System: Docker
• Channels: Telegram

📸 Steps to Reproduce

1. Add a groq API key to model_list section
2. Also add a groq API key to deprecated providers section
3. Send a voice message via Telegram to picoclaw

❌ Actual Behavior

• Picoclaw only sees [voice]

✅ Expected Behavior

• Transcriber should kick in

💬 Additional Context

[nikolasdehor]

Confirmed. The voice.GroqTranscriber exists in pkg/voice/transcriber.go but is never instantiated or called anywhere in the codebase. Here is exactly what happens:

Telegram voice path (pkg/channels/telegram/telegram.go:480-487):

voicePath := c.downloadFile(ctx, message.Voice.FileID, ".ogg")
if voicePath != "" {
    mediaPaths = append(mediaPaths, storeMedia(voicePath, "voice.ogg"))
}
// ...
content += "[voice]"

The voice OGG file is downloaded and stored as a media reference, but the content sent to the LLM is just the literal string "[voice]". No transcription happens.

Same pattern in other channels:

• OneBot (pkg/channels/onebot/onebot.go:807): downloads voice.amr, stores it, but no transcription
• WeCom (pkg/channels/wecom/): similar pattern

What is missing:

1. Nobody calls voice.NewGroqTranscriber(apiKey) - grep across the entire codebase shows it is only defined in pkg/voice/transcriber.go, never imported or used
2. No transcription step exists between downloading the audio file and publishing the inbound message
3. The Groq API key for transcription is not resolved from the new model_list config format

Where the wiring should go:

The transcription should happen in each channel after downloading the voice file and before publishing the inbound message. For Telegram, around line 480-487:

voicePath := c.downloadFile(ctx, message.Voice.FileID, ".ogg")
if voicePath != "" {
    mediaPaths = append(mediaPaths, storeMedia(voicePath, "voice.ogg"))
    if transcriber != nil && transcriber.IsAvailable() {
        resp, err := transcriber.Transcribe(ctx, voicePath)
        if err == nil && resp.Text != "" {
            content += resp.Text
        } else {
            content += "[voice: transcription failed]"
        }
    } else {
        content += "[voice]"
    }
}

The GroqTranscriber needs to be injected into each channel that handles voice messages (via constructor parameter or a shared service). The API key resolution should check model_list for a groq provider entry.

This looks like it regressed during the v0.2.0 refactor when the provider initialization was moved to model_list - the old code likely had the transcriber wired through the legacy providers config path.